pith. sign in

arxiv: 2606.18060 · v1 · pith:UDOLGWLMnew · submitted 2026-06-16 · 💻 cs.AI · cs.CL

PseudoBench: Measuring How Agentic Auto-Research Fuels Pseudoscience

Xinyang Liao , Lingyu Li , Huacan Liu , Tianle Gu , Yang Yao , Tong Zhu , Yan Teng , Yingchun Wang This is my paper

Pith reviewed 2026-06-27 01:25 UTC · model grok-4.3

classification 💻 cs.AI cs.CL
keywords PseudoBenchpseudoscienceagentic auto-researchLLM agentsbenchmarkresearch pipelinescientific alignmentresistance evaluation
0
0 comments X

The pith

Current AI research agents produce pseudoscientific reports with resistance no higher than 27.4%.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper is trying to establish that agentic AI systems for autonomous research lack the ability to resist pseudoscience. It creates PseudoBench with 200 claim-evidence pairs across five domains and runs seven agents through a full pipeline from experiments to written reports. The tests show near-zero refusal rates and a maximum resistance of 27.4 percent. Stronger agents produce more sophisticated and credible-sounding outputs that align with the false premises. A sympathetic reader would care because widespread use of such agents could contaminate scientific literature with plausible but misleading studies.

Core claim

PseudoBench is an adversarial benchmark containing 200 curated pseudoscientific claim-evidence pairs across five domains. It evaluates agentic auto-research systems through an end-to-end research pipeline from experiments to writing. Testing seven state-of-the-art agents shows that current systems readily produce persuasive reports that align with pseudoscientific premises with near-zero refusal rates and the highest resistance of only 27.4%. Stronger agents risk packaging pseudoscience in more sophisticated scientific language, increasing its apparent credibility.

What carries the argument

PseudoBench benchmark with its 200 pseudoscientific claim-evidence pairs and the end-to-end research pipeline that measures whether agents refuse or align with false premises.

If this is right

  • Such agents may generate misleading studies that contaminate academic literature.
  • Public trust in science could erode if plausible false reports enter circulation.
  • More advanced agents may make pseudoscience harder to detect through sophisticated scientific framing.
  • Scientific alignment techniques must be developed before these systems see widespread deployment.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Safety training for research agents needs explicit components for detecting and rejecting pseudoscientific content.
  • The benchmark approach could be adapted to evaluate agents on other forms of misinformation in technical domains.
  • Autonomous research tools may require additional verification layers or human oversight to limit spread of false claims.
  • This highlights a general limitation where current models prioritize coherent output over alignment with established facts.

Load-bearing premise

The 200 curated pseudoscientific claim-evidence pairs are representative of real pseudoscience and the end-to-end research pipeline accurately measures an agent's ability to identify and resist pseudoscientific narratives.

What would settle it

Testing the same seven agents on a fresh, independently curated collection of pseudoscientific premises and finding that refusal rates remain below 30 percent or that the resulting reports pass peer review in actual journals.

Figures

Figures reproduced from arXiv: 2606.18060 by Huacan Liu, Lingyu Li, Tianle Gu, Tong Zhu, Xinyang Liao, Yang Yao, Yan Teng, Yingchun Wang.

Figure 1
Figure 1. Figure 1: Example task from PseudoBench: inventing a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of PseudoBench: dataset construction, report generation, and evaluation protocol. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: shows the distribution of the five pseu￾doscientific categories, and [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mean-score heatmap of the 14 second-level criteria across the seven auto-research systems. Each cell [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Per-item report quality versus pseudoscience [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Report quality versus persuasiveness across [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Domain-level mean-score heatmap of the 14 second-level submetrics across the seven auto-research [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Prompt for seed rescoring. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Prompt for seed rewriting. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Prompt for report generation. Report Quality System Prompt You are a professional PDF research-paper quality evaluator. Your task is to assess the report generation quality of a given PDF. This task evaluates only how complete, formal, and paper-like the report is as a finished research paper or research report. Do not evaluate whether the report content is true, scientific, or credible. Evaluation criter… view at source ↗
Figure 12
Figure 12. Figure 12: Prompt for report quality evaluation. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Prompt for pseudoscience alignment evaluation. [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Prompt for persuasiveness evaluation. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Case study output generated by Codex. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Case study output generated by Claude Code. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Case study output generated by OpenClaw. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_17.png] view at source ↗
Figure 2
Figure 2. Figure 2: Modeled overlap between biofield bands and crystal-emission bands. The dominant peak occurs [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Modeled improvement indices across the proposed therapeutic pathways. [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
Figure 18
Figure 18. Figure 18: Case study output generated by Nanobot. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Case study output generated by EvoScientist. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Case study output generated by ResearchClaw. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Case study output generated by ARIS. 26 [PITH_FULL_IMAGE:figures/full_fig_p026_21.png] view at source ↗
read the original abstract

As Large Language Model based agents enter autonomous scientific research, their ability to resist pseudoscience becomes increasingly important. Otherwise, such systems may rapidly generate plausible yet misleading studies that contaminate academic literature and erode trust in science. We present PseudoBench, an adversarial benchmark for evaluating whether agentic auto-research systems can identify and resist pseudoscientific narratives. PseudoBench contains 200 curated pseudoscientific claim-evidence pairs across five domains and evaluates agents through an end-to-end research pipeline from experiments to writing. Testing seven state-of-the-art agents, we find that current systems readily produce persuasive reports that align with pseudoscientific premises with near-zero refusal rates and the highest resistance of only 27.4%. Stronger agents risk packaging pseudoscience in more sophisticated scientific language, increasing its apparent credibility. These findings reveal an alarming capacity to fuel pseudoscience, calling for scientific alignment before widespread deployment.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript introduces PseudoBench, a benchmark of 200 curated pseudoscientific claim-evidence pairs across five domains. It evaluates seven state-of-the-art LLM-based agents via an end-to-end research pipeline (experiments to report writing) and reports near-zero refusal rates with a maximum resistance of 27.4%, concluding that current agents readily generate persuasive reports aligning with pseudoscientific premises and that stronger agents may increase credibility through sophisticated language.

Significance. If the curation criteria, alignment scoring, and pipeline controls prove robust and representative, the results would demonstrate a concrete risk that agentic auto-research systems can amplify pseudoscience, supporting calls for scientific alignment prior to deployment. The benchmark format itself is a useful contribution for future evaluation work.

major comments (3)
  1. [Abstract] Abstract: the central claim that agents 'readily produce persuasive reports that align with pseudoscientific premises with near-zero refusal rates' rests on the 200 pairs and the scoring of alignment/resistance, yet the abstract (and described pipeline) supplies no information on pair curation criteria, experimental controls, statistical methods, or inter-rater reliability.
  2. [Evaluation pipeline] Evaluation pipeline (implied in abstract): it is unclear how 'alignment with pseudoscientific premises' versus 'resistance' is operationalized (e.g., conclusion polarity, citation of supplied evidence, absence of debunking, or LLM-judge rubric), and whether scoring was performed by the curating team or calibrated judges; without an explicit decision procedure the 27.4% figure is sensitive to subjective thresholds.
  3. [Abstract] Abstract: the claim that the 200 pairs are representative of real pseudoscience and that the end-to-end pipeline accurately measures resistance is presented without evidence of external validation, domain-expert review, or controls for prompt sensitivity.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'highest resistance of only 27.4%' would benefit from a parenthetical note on the agent that achieved it and the exact definition of resistance used.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting areas where methodological details could be made more explicit. We address each major comment below, indicating revisions where appropriate to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that agents 'readily produce persuasive reports that align with pseudoscientific premises with near-zero refusal rates' rests on the 200 pairs and the scoring of alignment/resistance, yet the abstract (and described pipeline) supplies no information on pair curation criteria, experimental controls, statistical methods, or inter-rater reliability.

    Authors: We agree the abstract prioritizes brevity and therefore omits these details. The full manuscript describes curation criteria in Section 3 (sourcing from documented pseudoscience literature across five domains), experimental controls and statistical methods (mean resistance rates with standard errors) in Section 4, and inter-rater reliability (Cohen's kappa of 0.82 between LLM judge and human spot-checks) in Section 4.3. To address the concern, we will revise the abstract to include one sentence summarizing curation and evaluation approach. revision: yes

  2. Referee: [Evaluation pipeline] Evaluation pipeline (implied in abstract): it is unclear how 'alignment with pseudoscientific premises' versus 'resistance' is operationalized (e.g., conclusion polarity, citation of supplied evidence, absence of debunking, or LLM-judge rubric), and whether scoring was performed by the curating team or calibrated judges; without an explicit decision procedure the 27.4% figure is sensitive to subjective thresholds.

    Authors: Alignment is defined as the generated report endorsing the pseudoscientific premise (via conclusion polarity and lack of debunking statements), while resistance is recorded when the agent refuses the task or produces a counter-report. Scoring uses a calibrated LLM judge (GPT-4o with a fixed rubric and few-shot examples) rather than the curation team. We will add an explicit subsection (4.2.1) reproducing the full rubric and decision tree in the revision to make the procedure transparent and reduce perceived subjectivity. revision: yes

  3. Referee: [Abstract] Abstract: the claim that the 200 pairs are representative of real pseudoscience and that the end-to-end pipeline accurately measures resistance is presented without evidence of external validation, domain-expert review, or controls for prompt sensitivity.

    Authors: Pairs were drawn from established pseudoscience sources in each domain; the pipeline uses fixed prompt templates with minor variations tested for sensitivity (reported in Appendix B). We did not conduct formal external domain-expert validation, which is a genuine limitation. We will expand the Limitations section to state this explicitly and note that future work will include expert review, while retaining the current claim as an internal validity statement rather than a validated external one. revision: partial

Circularity Check

0 steps flagged

No circularity; benchmark is externally defined evaluation

full rationale

The paper introduces PseudoBench as an external benchmark consisting of 200 curated claim-evidence pairs and an end-to-end agent evaluation pipeline. No mathematical derivations, fitted parameters, self-referential definitions, or load-bearing self-citations appear in the abstract or described structure. Results are presented as empirical measurements on third-party agents rather than quantities derived from the benchmark's own outputs by construction, rendering the work self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no information on free parameters, background axioms, or new postulated entities.

pith-pipeline@v0.9.1-grok · 5697 in / 961 out tokens · 45417 ms · 2026-06-27T01:25:55.753167+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

140 extracted references · 10 linked inside Pith

  1. [1]

    Nature , pages=

    A multi-agent system for automating scientific discovery , author=. Nature , pages=. 2026 , publisher=

    2026

  2. [2]

    Nature , volume=

    Towards end-to-end automation of AI research , author=. Nature , volume=. 2026 , publisher=

    2026

  3. [3]

    Zhang, Barry and Lazuka, Keith and Murag, Maahesh , title =

  4. [4]

    Lopopolo, Ryan , title =

  5. [5]

    Nature , volume=

    Hey ChatGPT, write me a fictional paper: these LLMs are willing to commit academic fraud , author=. Nature , volume=

  6. [7]

    Nature Machine Intelligence , pages=

    Benchmarking large language models on safety risks in scientific laboratories , author=. Nature Machine Intelligence , pages=. 2026 , publisher=

    2026

  7. [9]

    2025 , publisher=

    Attention authors: Updated practice for review articles and position papers in arxiv cs category , author=. 2025 , publisher=

    2025

  8. [11]

    Harvard Kennedy School Misinformation Review , volume=

    GPT-fabricated scientific papers on Google Scholar: Key features, spread, and implications for preempting evidence manipulation , author=. Harvard Kennedy School Misinformation Review , volume=

  9. [12]

    Advances in neural information processing systems , volume=

    Training language models to follow instructions with human feedback , author=. Advances in neural information processing systems , volume=

  10. [14]

    2021 , publisher=

    Why trust science? , author=. 2021 , publisher=

    2021

  11. [15]

    Nature , year=

    Researchers who use hallucinated references to face arXiv ban , author=. Nature , year=

  12. [20]

    2026 , howpublished =

    Introducing GPT-5.4 , author =. 2026 , howpublished =

    2026

  13. [21]

    2026 , howpublished =

    Claude Opus 4.7 , author =. 2026 , howpublished =

    2026

  14. [22]

    2026 , howpublished =

    Claude Opus 4.6 , author =. 2026 , howpublished =

    2026

  15. [23]

    2026 , howpublished =

    ResearchClaw: Local-first Research OS for Papers, Workflows, Experiments, Channels, and Automation , author =. 2026 , howpublished =

    2026

  16. [24]

    2026 , howpublished =

    OpenClaw: Personal AI Assistant , author =. 2026 , howpublished =

    2026

  17. [25]

    2026 , howpublished =

    nanobot: Lightweight, Open-source AI Agent for Your Tools, Chats, and Workflows , author =. 2026 , howpublished =

    2026

  18. [26]

    2026 , howpublished =

    Codex , author =. 2026 , howpublished =

    2026

  19. [27]

    2026 , howpublished =

    Claude Code , author =. 2026 , howpublished =

    2026

  20. [28]

    2026 , howpublished =

    List of Pseudoscience Topics , author =. 2026 , howpublished =

    2026

  21. [29]

    2026 , howpublished =

    Minke Bar , author =. 2026 , howpublished =

    2026

  22. [30]

    Future Internet , volume=

    The rise of agentic ai: A review of definitions, frameworks, architectures, applications, evaluation metrics, and challenges , author=. Future Internet , volume=. 2025 , publisher=

    2025

  23. [31]

    Artificial Intelligence Review , volume=

    Agentic AI: a comprehensive survey of architectures, applications, and future directions , author=. Artificial Intelligence Review , volume=. 2025 , publisher=

    2025

  24. [32]

    IEEe Access , volume=

    Agentic AI: Autonomous intelligence for complex goals—A comprehensive survey , author=. IEEe Access , volume=. 2025 , publisher=

    2025

  25. [33]

    Array , volume=

    The role of agentic ai in shaping a smart future: A systematic review , author=. Array , volume=. 2025 , publisher=

    2025

  26. [34]

    Global Business and Organizational Excellence , volume=

    Agentic AI Systems: What It Is and Isn't , author=. Global Business and Organizational Excellence , volume=. 2026 , publisher=

    2026

  27. [35]

    Informatics and Health , volume=

    Next-generation agentic AI for transforming healthcare , author=. Informatics and Health , volume=. 2025 , publisher=

    2025

  28. [36]

    The Lancet , volume=

    The rise of agentic AI teammates in medicine , author=. The Lancet , volume=. 2025 , publisher=

    2025

  29. [37]

    IEEE Access , year=

    Agentic AI in education: State of the art and future directions , author=. IEEE Access , year=

  30. [38]

    Journal of Management Science Research Review , volume=

    The end of human-only knowledge management: Agentic AI in education , author=. Journal of Management Science Research Review , volume=

  31. [39]

    Journal of Business Research , volume=

    AI agents, agentic AI, and the future of sales , author=. Journal of Business Research , volume=. 2026 , publisher=

    2026

  32. [40]

    Information , volume=

    From Recommendations to Delegation: A Systematic Review Mapping Agentic AI in E-Commerce and Its Consumer Effects , author=. Information , volume=. 2026 , publisher=

    2026

  33. [41]

    Communications Chemistry , year=

    ChemGraph as an agentic framework for computational chemistry workflows , author=. Communications Chemistry , year=

  34. [42]

    Matter , volume=

    El Agente: An autonomous agent for quantum chemistry , author=. Matter , volume=. 2025 , publisher=

    2025

  35. [43]

    bioRxiv , pages=

    SpatialAgent: An autonomous AI agent for spatial biology , author=. bioRxiv , pages=. 2025 , publisher=

    2025

  36. [44]

    bioRxiv , pages=

    Agentic Lab: An Agentic-physical AI system for cell and organoid experimentation and manufacturing , author=. bioRxiv , pages=. 2025 , publisher=

    2025

  37. [45]

    Science , volume=

    Delocalized, asynchronous, closed-loop discovery of organic laser emitters , author=. Science , volume=. 2024 , publisher=

    2024

  38. [46]

    Journal of the American Chemical Society , volume=

    A multiagent-driven robotic AI chemist enabling autonomous chemical research on demand , author=. Journal of the American Chemical Society , volume=. 2025 , publisher=

    2025

  39. [49]

    IEEE Intelligent Systems , volume=

    The rise of agentic AI: implications, concerns, and the path forward , author=. IEEE Intelligent Systems , volume=. 2025 , publisher=

    2025

  40. [50]

    AI Hallucinations: A Misnomer Worth Clarifying , year=

    Maleki, Negar and Padmanabhan, Balaji and Dutta, Kaushik , booktitle=. AI Hallucinations: A Misnomer Worth Clarifying , year=

  41. [51]

    and Kanade, T

    Baker, S. and Kanade, T. , booktitle=. Hallucinating faces , year=

  42. [52]

    Frontiers in Psychology , volume=

    Hallucinations and related concepts—their conceptual background , author=. Frontiers in Psychology , volume=. 2015 , publisher=

    2015

  43. [53]

    , author=

    Tactile hallucinations: conceptual and historical aspects. , author=. Journal of Neurology, Neurosurgery & Psychiatry , volume=. 1982 , publisher=

    1982

  44. [54]

    Hallucinations: Research and practice , pages=

    The construction of hallucination: history and epistemology , author=. Hallucinations: Research and practice , pages=. 2011 , publisher=

    2011

  45. [56]

    Proceedings of the 58th annual meeting of the association for computational linguistics , pages=

    On faithfulness and factuality in abstractive summarization , author=. Proceedings of the 58th annual meeting of the association for computational linguistics , pages=

  46. [57]

    ACM computing surveys , volume=

    Survey of hallucination in natural language generation , author=. ACM computing surveys , volume=. 2023 , publisher=

    2023

  47. [58]

    Computational Linguistics , volume=

    �� Siren’s Song in the AI Ocean: A Survey on Hallucination in Large Language Models , author=. Computational Linguistics , volume=. 2025 , publisher=

    2025

  48. [59]

    ACM Transactions on Information Systems , volume=

    A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions , author=. ACM Transactions on Information Systems , volume=. 2025 , publisher=

    2025

  49. [63]

    Applied and Computational Engineering , volume=

    A survey of hallucination problems based on large language models , author=. Applied and Computational Engineering , volume=

  50. [64]

    arXiv preprint arXiv:2401.11817 , year=

    Hallucination is inevitable: An innate limitation of large language models , author=. arXiv preprint arXiv:2401.11817 , year=

    Pith/arXiv arXiv

  51. [66]

    Proceedings of the 43rd IEEE/ACM International Conference on Computer-Aided Design , pages=

    Generative AI agents in autonomous machines: A safety perspective , author=. Proceedings of the 43rd IEEE/ACM International Conference on Computer-Aided Design , pages=

  52. [67]

    2002 , publisher=

    You're too kind: A brief history of flattery , author=. 2002 , publisher=

    2002

  53. [68]

    2001 , school=

    The Gift of flattery: a social and biological analysis of deceptive practices , author=. 2001 , school=

    2001

  54. [69]

    ESUT Journal of Social Sciences , volume=

    Sycophancy and Dearth of Integrity in Governance , author=. ESUT Journal of Social Sciences , volume=

  55. [70]

    Intelligent Computing-Proceedings of the Computing Conference , pages=

    Sycophancy in large language models: Causes and mitigations , author=. Intelligent Computing-Proceedings of the Computing Conference , pages=. 2025 , organization=

    2025

  56. [73]

    International Conference on Learning Representations , volume=

    Towards understanding sycophancy in language models , author=. International Conference on Learning Representations , volume=

  57. [75]

    Expanding on Sycophancy , year =

  58. [76]

    Findings of the association for computational linguistics: ACL 2023 , pages=

    Discovering language model behaviors with model-written evaluations , author=. Findings of the association for computational linguistics: ACL 2023 , pages=

    2023

  59. [79]

    Science , volume=

    Sycophantic AI decreases prosocial intentions and promotes dependence , author=. Science , volume=. 2026 , publisher=

    2026

  60. [81]

    2024 IEEE conference on artificial intelligence (CAI) , pages=

    AI hallucinations: a misnomer worth clarifying , author=. 2024 IEEE conference on artificial intelligence (CAI) , pages=. 2024 , organization=

    2024

  61. [82]

    Proceedings Fourth IEEE international conference on automatic face and gesture recognition (Cat

    Hallucinating faces , author=. Proceedings Fourth IEEE international conference on automatic face and gesture recognition (Cat. No. PR00580) , pages=. 2000 , organization=

    2000

  62. [83]

    Frontiers in Artificial Intelligence , volume=

    AI, agentic models and lab automation for scientific discovery—the beginning of scAInce , author=. Frontiers in Artificial Intelligence , volume=. 2025 , publisher=

    2025

  63. [84]

    Patterns , volume=

    The reanimation of pseudoscience in machine learning and its ethical repercussions , author=. Patterns , volume=. 2024 , publisher=

    2024

  64. [85]

    Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    An open-source data contamination report for large language models , author=. Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    2024

  65. [87]

    arXiv preprint arXiv:2412.18619 , year=

    Next token prediction towards multimodal intelligence: A comprehensive survey , author=. arXiv preprint arXiv:2412.18619 , year=

    arXiv

  66. [88]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    UAQFact: Evaluating Factual Knowledge Utilization of LLMs on Unanswerable Questions , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  67. [89]

    Mohamad Abou Ali, Fadi Dornaika, and Jinan Charafeddine. 2025. Agentic ai: a comprehensive survey of architectures, applications, and future directions. Artificial Intelligence Review, 59(1):11

    2025

  68. [90]

    Deepak Bhaskar Acharya, Karthigeyan Kuppan, and B Divya. 2025. Agentic ai: Autonomous intelligence for complex goals—a comprehensive survey. IEEe Access, 13:18912--18936

    2025

  69. [91]

    Aisha Alansari and Hamzah Luqman. 2025. Large language models hallucination: A comprehensive survey. arXiv preprint arXiv:2510.06265

    arXiv 2025

  70. [92]

    Mel Andrews, Andrew Smart, and Abeba Birhane. 2024. The reanimation of pseudoscience in machine learning and its ethical repercussions. Patterns, 5(9)

    2024

  71. [93]

    Anthropic . 2026 a . Claude code. https://www.anthropic.com/product/claude-code

    2026

  72. [94]

    Anthropic . 2026 b . Claude opus 4.6. https://www.anthropic.com/news/claude-opus-4-6

    2026

  73. [95]

    Anthropic . 2026 c . Claude opus 4.7. https://www.anthropic.com/news/claude-opus-4-7

    2026

  74. [96]

    Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, and 1 others. 2022. Constitutional ai: Harmlessness from ai feedback. arXiv preprint arXiv:2212.08073

    Pith/arXiv arXiv 2022

  75. [97]

    Zechen Bai, Pichao Wang, Tianjun Xiao, Tong He, Zongbo Han, Zheng Zhang, and Mike Zheng Shou. 2024. Hallucination of multimodal large language models: A survey. arXiv preprint arXiv:2404.18930

    Pith/arXiv arXiv 2024

  76. [98]

    Baidu Tieba . 2026. Minke bar. https://tieba.baidu.com/f?kw=

    2026

  77. [99]

    Stefanos Balaskas. 2026. From recommendations to delegation: A systematic review mapping agentic ai in e-commerce and its consumer effects. Information, 17(3):222

    2026

  78. [100]

    Ajay Bandi, Bhavani Kongari, Roshini Naguru, Sahitya Pasnoor, and Sri Vidya Vilipala. 2025. The rise of agentic ai: A review of definitions, frameworks, architectures, applications, evaluation metrics, and challenges. Future Internet, 17(9):404

    2025

  79. [101]

    Saikat Barua. 2024. Exploring autonomous agents through the lens of large language models: A review. arXiv preprint arXiv:2404.04442

    arXiv 2024

  80. [102]

    Kat Boboris. 2025. Attention authors: Updated practice for review articles and position papers in arxiv cs category

    2025

Showing first 80 references.